Big Blue vs. Green Giant in a fight for science’s most coveted exascale workloads.
Intel is, first and foremost, a computing business – it's the company's first priority to sell as many processors as possible to as many customers as possible.
While they've had outstanding success in the traditional CPU market, Intel have been eyeing off a market segment that until recently has remained relatively untouched by the big blue: highly-parallel smaller-scale supercomputing.
Announced on June 20 2010 were plans to reincarnate the unsuccessful GPU project Larrabee as a reinvigorated supercomputing platform, known as Knight's Corner. It would be manufactured on a 22nm process, and feature 50 x86 processing cores, enabling high amounts of massively-threaded data throughput.
Exactly a year later came an announcement that first-generation Knight's Corner product, Aubrey Isle, will finally be thrown into full-scale production under the umbrella term MIC – short for Many Integrated Core architecture.
Intel Vice President Kirk Skaugen highlighted the company's plan for Knight's Corner: "While Intel® Xeon® processors are the clear architecture of choice for the current TOP500 list of supercomputers, Intel is further expanding its focus on high-performance computing by enabling the industry for the next frontier with our Many Integrated Core architecture for petascale and future exascale workloads."
This announcement has implications for the M2090, NVIDIA's Tesla product, a somewhat similar supercomputing platform that instead of using traditional x86 cores, uses 512 graphics processing cores. The downsides of this platform are that it's highly specialised and code must be ported to the CUDA language; however the performance is certainly attractive.
Intel's hoping to grab a slice of the NVIDIA supercomputing pie – which notably made recent breakthroughs in China's solar research – tempting customers with a platform that can run x86 code somewhat more easily.
Performance of Knight's Corner was not announced, though we speculate it will be less useful for massively-threaded applications and more useful for largely-threaded workloads. Either way, Knight's Corner will form part of the next decade's scientific advances – whether NVIDIA likes it or not.